Emulation of expensive simulation model for operation and maintenance of offshore wind farms

As wind farms move deeper offshore to tap into stronger and relentless winds, intense wind and wave conditions pose a great challenge in terms of their operation and maintenance (O&M). There are several factors that determine the profitability of offshore wind farms, and the most critical factors among them are the parameters on allocation of maintenance resources. These parameters interact with environmental factors and make it impossible to estimate profitability using simple formulas. On the other hand, existing simulation models, which describe the behaviour of wind farms by using mathematical models of wind, wave, and their effects on O&M, can be extremely detailed resulting in simulations being computationally very expensive. Depending on the number of scenarios to evaluate, it can take up-to several days to complete the computation. In order to address this difficulty, a statistical model fitting approach has been adopted to emulate the behaviour of the computationally expensive simulator. Neural networks, splines, and decision trees are combined to capture numerical and discrete variables and their influence on availability and profitability. This approach is useful because it allows for quick exploration of the space of operating choices, which would be difficult to achieve by repeated simulations due to their computational expense. The performance results show that the statistical model can evaluate hundreds of scenarios per second, and the approximation error is acceptable.